TT

The Trust Library API

This library has been developed as a proof of concept concerning the use of trust in on line applications. Unlike most discussions of trust, this is not a heuristic cryptotoken or credential scheme. Rather it develops the idea that trust is a running assessment of a relationship between observer and observed in a system.

The code presented here is free to use, but it is to be thought of as a proof of concept, since much of it is tailored to the specific use cases for which data could be obtained. The code has been purposely designed with few entry points:

The purpose of any machine learning scheme is to condense inputs to provide a dimensional reduction of the original data.

Data types and SST context

The library makes use of the Semantic Spacetime model context, which has supporting infrastructure currently based on ArangoDB.

A Go application using the library can open an Analytics context as follows:

    TT.InitializeSmartSpaceTime()

    var dbname string = "SST-ML"
    var dburl string = "http://localhost:8529" // ArangoDB port
    var user string = "root"
    var pwd string = "mark"

    G = TT.OpenAnalytics(dbname,dburl,user,pwd)

Transaction wrappers

Two sets of functions for wrapping transactional events or critical sections parenthetically (with begin-end semantics).

• Functions that timestamp transactions at the moment of measurement according to the local system clock

 PromiseContext_Begin(g Analytics, name string) PromiseContext 
 PromiseContext_End(g Analytics, ctx PromiseContext) PromiseHistory 

• Functions with Golang time stamp supplied from outside, e.g. for offline analysis with epoch timestamps.

 StampedPromiseContext_Begin(g Analytics, name string, before time.Time) PromiseContext 
 StampedPromiseContext_End(g Analytics, ctx PromiseContext, after time.Time) PromiseHistory

Periodic Key Value Storage

Most time based data can be classified according to their point in a week from Monday to Friday with finite resolution of five minute intervals. Most system interactions have correlation times of up to 20 minutes. There is no need to collect nanosecond metrics for functioning systems. Each collection name (collname) represents a vector of values indexed by Unix or Golang timestamps which are turned into 5 minute intervals.

These functions fail silently with empty data rather than raising internal errors. The main reason for failure is incorrect naming.

 SumWeeklyKV(g Analytics, collname string, t int64, value float64)
 LearnWeeklyKV(g Analytics, collname string, t int64, value float64)

 AddWeeklyKV_Unix(g Analytics, collname string, t int64, value float64)
 AddWeeklyKV_Go(g Analytics, collname string, t time.Time, value float64)

To retrieve data from the learning store ordered by the working week,

 GetAllWeekMemory(g Analytics, collname string) []float64 

Text n-gram analysis

To scan a body of text, we first have to strip is clear of any encoding. Examples for handling UNICODE UTF-8 are given in the Chinese language n-gram example. One first selects an appropriate paragraph size for the kind of text one is analysing. For book text or narrative, a value of 100 is okay; for short notes a value of 10 is more appropriate. The learning radius determines the approximate amount of text that one considers to be coherently about the same thing. It affects the amount of the whole that gets subsampled during summarization.

	const paragraph_radius = 100
	return TT.TextPrism(subject, text, paragraph_radius)

In verbose mode, these generate a lot of helpful output to help understand the analysis.

Heuristics context

The TT library also contains a heuristic symbol evaluator, CFEngine style. This remains a simple lightweight approach more powerful than several others, and does a similar job to Open Policy Agent, etc. The interface is deliberately y simple.

-InitializeContext() - Reset and empty the set of active context classes

-ContextAdd(s string) - Add a symbol idempotently to the context set

-ContextSet() []string - Return an array of context symbols

-Context(expression string) bool - Evaluate whether the expression is true or false in terms of its heuristic score

Example:

TT.InitializeContext()

ContextAdd("busy")

if TT.Context("busy || lazy && slow") {
  // policy ...
}

fmt.Println("Context:",TT.ContextSet())